I'm working on a study that involves predicting a surgery binary outcome (bailout yes vs. bailout no) in a hospital patient dataset. We wanted to include surgeon-level variables in a mixed model and we calculated the outcome rate per surgeon as one of the surgeon-level variables. For example we can find out from the dataset a specific surgeon has a bailout rate of 70% from the number of surgeries they performed in the dataset. And that surgeon would have 0.7 as the value of the surgeon bailout rate variable for all of his surgeries. I'm wondering if it's okay to input this "surgeon bailout rate" variable as a predictor for "bailout" given that this variable is derived from the outcome. Any thoughts appreciated. Thanks.
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ Sounds like you want binomial regression. Bailing out of 7/10 is not the same as bailing out of 700/1000. $\endgroup$– Jeremy MilesCommented Jun 4, 2021 at 16:12
-
1$\begingroup$ How many surgeons are involved? Are surgeons treated as the random effects in the mixed model? Do you want to be able to distinguish among individual surgeons in terms of bailout rate, or do you just want an estimate of the variability among surgeons? How many other/which predictors are involved in your model? How are you proposing to use the model? Please provide that information by editing the question, as comments are easily overlooked and can get lost. Also, please define "bailout" in this context. $\endgroup$– EdMCommented Jun 4, 2021 at 16:20
Add a comment
|